OpenAirInterface promoting the development, distribution and adoption of the opesource hardware and software wireless technology platforms

Size: px

Start display at page:

Download "OpenAirInterface promoting the development, distribution and adoption of the opesource hardware and software wireless technology platforms"

Susan Robbins
6 years ago
Views:

1 OpenAirInterface promoting the development, distribution and adoption of the opesource hardware and software wireless technology platforms Navid Niakein EURECOM, Mobile Communication Department This work is licensed under a CC attribution Share-Alike 3.0 Unported license.

2 OpenAirInterface in a Nutshell Open-source (hardware and software) wireless technology platforms for deployment of mock network with high level of realism Soft modem: SDR architecture and full GPP System approach with high level of realism Part of FIRE facility: remote access Current focus 3GPP LTE (unicast and multicast), and a subset of LTE-A features IEEE p and LTE meshing extension Objectives Open forum for innovation in air-interface technologies and wireless networking through experimentation Open platforms and proof-of-concepts through real-time prototypes and scalable emulation platforms Dissemination, education, and training, project - p 2

OAI Wireless technology Platforms Simulation/ Emulation Link-level System-level Emulation Field trail Experimentation & Innovations Soft Modem Target: x86 arch EXMIMO 2 USRP EMOS Simulation FPGA

3 OAI Wireless technology Platforms Simulation/ Emulation Link-level System-level Emulation Field trail Experimentation & Innovations Soft Modem Target: x86 arch EXMIMO 2 USRP EMOS Simulation FPGA Modem EXMIMO I SoC Arch. C implementation under realtime Linux optimized for x86 Development and integration methodology Tight interaction between the system emulation and soft modem Rich R&D environment: Aeroflex-Geisler LEON/GRLIB, RTAI, Linux, GNU, Wireshark, control and monitoring tools, message and time analyzer, Low-level log processing, traffic generator, profiling tools, soft scope. - p 3

IP stack NAS enb Application NAS HSS S11 S1-U RRC RRC S1-MME X2AP S1-U S1-MME S6a/Diameter GTP-U SGi PDCP PDCP SCTP UDP

4 OpenAirInterface.org Access Stratum + EPCLite GNU GPL License IP packets UE enb EPC AT commands MME Application S+P-GW Application Linux IP stack NAS enb Application NAS HSS S11 S1-U RRC RRC S1-MME X2AP S1-U S1-MME S6a/Diameter GTP-U SGi PDCP PDCP SCTP UDP SCTP UDP RLC RLC IP IP MAC MAC Ethernet Ethernet PHY PHY UEs enbs MME + S+P-GW 3GPP layers Linux stack Control Plane Data Plane IoT OAI enb with a UE dongle (Huawei E392U-12) OAI enb with smartphone OAI enb with other EPC - p

5 Hardware Targets for OpenairG Designed and maintained by EURECOM Used by many academia/industrial partners 1.5/5/10/20 MHz, FDD/TDD (MIMO) USRP B210 Commercial Ettus/National Instruments board Software drivers originally maintained by Orange Labs (Beijing), now officially supported by Eurecom Not 100% LTE compliant (limited to 6.5Msps) USRP X300 Coming soon - p 5

6 Express MIMO 2 RF RX ( way) RF TX ( way) PCI Express (1 or way) Spartan 6 LX150T xlms6002d RF ASICs 12V from ATX power supply 250 MHz 3.8 GHz GPIO for external RF control - p 6

Software Architecture (LEON) HW Space User Space Application

USERSPACE/LIB Kernel Space Linux network driver (nas_driver.

openair2/nas/driver/cellular Using real-time Linux extension

(ltesoftmodem) Linux driver (openair_rf.

7 Software Architecture (LEON) HW Space User Space Application Control / Monitoring PCIexpress targets/arch/exmimo/ USERSPACE/LIB Kernel Space Linux network driver (nas_driver.ko) Openair2/NAS/DRIVER/MESH/ openair2/nas/driver/lite openair2/nas/driver/cellular Using real-time Linux extension (RTAI, Xenomai, RT-preemt) Modem control and sync. (ltesoftmodem) Linux driver (openair_rf.ko) targets/arch/exmimo/driver/eurecom targets/rtai/user Octave targets/arch/exmimo/userspace/octave - p 7

Emulator: in-lab system validation platform target: scalability Web Portal / Interface Scenario Results Scenario Descriptor Console Dispatcher Result Gen Input: Description of application scenario

8 Emulator: in-lab system validation platform target: scalability Web Portal / Interface Scenario Results Scenario Descriptor Console Dispatcher Result Gen Input: Description of application scenario Initialization and configuration of all the blocks Execution: PHY procedures, L2 protocols, traffic generator PHY abstraction, channel model, and mobility model Emulation medium: shared memory Output: Execution logs System/protocol operation Key performance indicators: latency, jitter, throughput/goodput Wireshark Log Gen. MSC & VCD Result Gen. Application Traffic Gen L3 Protocols OAI Network Interface L2 Protocols PHY Procedures PHY / PHY Abstraction Config. Gen. DRB Config. Traffic Gen. Ch. Realization Channel Model Path Loss Channel Descriptor Channel Trace Mobility Gen EMOS - p 8

9 CLOUDIFICATION OF OPENAIRINTERFACE - p 9

10 Cloud Radio Access Networks Two main steps : Soft RAN: to perform RAN functions on the top of general-purpose processor and not dedicated hardware such as DSP, FPGA, or ASIC Virtual RAN: to perform RAN functions on the top of virtualized cloud platforms sharing computing and storage capacity among different RAN functions - p 10

11 Soft RAN x86 Baseband DSP Challenge : efficient base band unit OpenAirInterface uses general-purpose x86 processors (GPP) for base-band processing front-end, channel decoding, phy procedures, L2 protocols Key elements Real-time extensions to Linux OS Real-time data acquisition to PC SIMD optimized integer DSP 6-bit MMX 128-bit SSE2/3/ 256-bit AVX2 Parallelism x86-6 : more efficient than legacy x86 - p 11

Soft RAN OAI BBU performance Three targets Intel IvyBridge i5-370 @ 3.

GHz small cell Configuration SISO MCS index: QPSK: 0,, 9 16QAM: 10, 13, 16 6QAM: 17, 22, 27

12 Soft RAN OAI BBU performance Three targets Intel IvyBridge 3.2GHz baseline Intel Xeon E C-RAN Intel Atom small cell Configuration SISO MCS index: QPSK: 0,, 9 16QAM: 10, 13, 16 6QAM: 17, 22, 27 PRB: 25 (5MHZ), 50 (10MHZ), 100 (20MHZ) gcc.7.3, x86-6,1000 frames, full rate, AWGN channel Not all the possible optimization are exploited - p 12

OAI BBU performance Intel IvyBridge i5-370 @ 3.

13 OAI BBU performance Intel IvyBridge 3.2GHz Timing(us) DL: PRB 25 UL: PRB 25 DL: PRB 50 UL: PRB 50 DL: PRB 100 UL: PRB 100 OAI BBU DL/UL vs MCS, Tx mode MCS Index - p 13

OAI BBU performance Intel Xeon E5-1607 v2@3ghz Timing(us) 2500 2000 1500 1000 DL: PRB 25 UL: PRB 25 DL: PRB

14 OAI BBU performance Intel Xeon E Timing(us) DL: PRB 25 UL: PRB 25 DL: PRB 50 UL: PRB 50 DL: PRB 100 UL: PRB 100 OAI BBU DL/UL vs MCS, Tx mode MCS Index - p 1

15 OAI BBU performance Intel Xeon E enb Rx stats (1subframe) OFDM demod : us ULSCH demod: us ULSCH Decoding : us 931 us (<1 core) enb Tx stats (1 subframe) OFDM mod : us DLSCH mod : us DLSCH scrambling : us DLSCH encoding : us 730 us (< 1core) With AVX2 (256-bit SIMD), turbo decoding and FFT processing will be exactly twice as fast 1 core per enb Configuration gcc.7.3, x86-6 (3 GHz Xeon E5-2690) 20 MHz bandwidth (UL mcs16 16QAM, DL mcs 2 6QAM, transmission mode 1 - SISO) 1000 frames, AWGN channel - p 15

16 OAI BBU performance Intel Atom Timing(us) DL: PRB 25 UL: PRB 25 DL: PRB 50 UL: PRB 50 DL: PRB 100 UL: PRB 100 OAI BBU DL/UL vs MCS, Tx mode MCS Index - p 16

17 Discussion With the possible/potential optimizations a full software solution would fit with an average of 1 x86 core per enb instance Mindful about the energy and realtime operation (potential existence of missing slots) using an external HW accelerators Reduce the soft modem computational complexity down to 0. cores per enb and improve energy efficiency Mindful about the bus utilization from the inner-modem to the hardware and back for further processing - p 17

18 Discussion BBU scales with average MCS: load balancing For a RRH MMMMMM ii < MMMMMM mmmmmm For a BBU MMMMMM NN MMMMMM mmmmmm Num CPU scales with nnnnnn CCCCCC MMMMMM 3 high-power RRH BBU Low-power or RX-only RRH - p 18

19 Parallelization of OAI BBU efficient use of available CPU cores and process prioritization Digital Baseband Inputs I[n] Q[n] Prefix Extraction FFT I[k] Q[k] Channel Compensation I[k] Q[k] LLR Unit Channel Decoding Highestpriority thread (scheduled every.5 ms) To timing (DAQ) and frequency correction units, and I/Q imbalance Pilot extraction Channel Estimation Broadband Channel Estimates lower-priority thread (scheduled every.5 ms) 8 lowestpriority threads (scheduled every 1 ms if needed) Slot 2n Slot 2n+1 Slot 2n+2 Slot 2n+3 CPU 3 RX DEC (2n) CPU 2 RX DEC (2n-2) CPU 1 RX LLR (2n-1) RX LLR (2n) RX LLR (2n+1) RX LLR (2n+2) CPU 0 TX (2n+2), RX (2n-1) RX (2n) TX (2n+), RX (2n+1) RX (2n+2) - p 19

From to RRH+CPRI Integration Today, boards support peak

5Gbps) 1eNB 20MHz 1Tx/1Rx (SISO) 1eNB 10MHz 2Tx/2Rx (MIMO)

commercial RRH with OpenAirInterface (future) Fronthaul RF

20 From to RRH+CPRI Integration Today, boards support peak throughput of 2Gbps (Theoretical PCI Express throughput: 2.5Gbps) 1eNB 20MHz 1Tx/1Rx (SISO) 1eNB 10MHz 2Tx/2Rx (MIMO) enb 5MHz 1Tx/1Rx (SISO) 1Core 1Core 1Core Connection of commercial RRH with OpenAirInterface (future) Fronthaul RF A/D D/A DAQ + CPRI OF CPRI (~0Gb Eth) CPRI + PCIe CPRI Switch 20

16-way PCIe Gen 2 (80 Gbit/s peak) High-End Intel Xeon Computing

21 Parallel use of various configurations are possible 16x 2.5Gbit/s =0 Gbit/s peak 6 RF + Antenna Ports 16-way PCIe Backplane 16-way PCIe Gen 2 (80 Gbit/s peak) High-End Intel Xeon Computing Engine 3GHz Dual-Proc AVX2 20 Parallel Cores OpenAirG RT-Linux MODEMs - p 21

22 OAI ECOSYSTEM - p 22

23 OAI use-cases and avenues supported by our past and on-going projects Standard and non-standard usage of 3GPP LTE systems Non-standard :IEEE802.21, PMIP and DMM, Exit native IP at the BS 5G evolution path of OAI soft-modem supported by EU/industrial projects Cloudification of radio networks (RAN+EPC) Massive MIMO, and COMP Cognitive networking Software-defined networking and network function virtualization support Support of machine type communications Mesh extension in support of multihop operation Cooperative transmission and MAC Caching strategy at the enb or S/P-GW Cooperative embms, proximity networking Scalable system experimentation and evaluation RRM policies, handover logic and performance, MIMO performance, traffic scheduling policy - p 23

24 Collaborative Web Tools OpenAirInterface SVN Repositories All development is available through s SVN repository (openairg) containing OPENAIR0 (open-source real-time HW/SW) OPENAIR1 (open-source real-time and offline SW) OPENAIR2 (open-source real-time and offline SW) OPENAIR3 (open-source Linux SW suite for cellular and MESH networks) TARGETS : different top-level target designs (emulator, RTAI, etc.) Partners can access and contribute to our development (RO access) OpenAirInterface TWIKI A TWIKI site for quick access by partners to our development via a collaborative HOW-TO Forum and bugzilla external support services (not currently used effectively) OAI VM image (start from openairg/targets/readme.txt) - p 2

Contacts Information URL: www.openairinterface.org https://twiki.eurecom.fr/twiki/bin/view/openairinterface Partnership and collaboration : openair_admin@eurecom.

25 Contacts Information URL: Partnership and collaboration : openair_admin@eurecom.fr Technical Support: openair_tech@eurecom.fr Developer support : openairg-devel@eurecom.fr To subscribe, send and to majordomo@eurecom.fr with the content "subscribe openairg-devel" - p 25

26 EURECOM MEMBERS - p 26

Wireless Radio Platforms

www.openairinterface.org Wireless Radio Platforms Mobile Communications Department Eurecom (Collaboration with SoC Laboratory, Telecom ParisTech Sophia) Facts OpenAirInterface Platform: Integrated RF/BB